Quantifying fish behavior in aquaculture using visual open-world tracking models

Monitoring and analyzing fish behaviors is critical for advancing aquaculture management and understanding fish welfare. Traditional methods, such as manual observation and sonar tracking, face significant limitations including inefficiency, low accuracy, and high dependence on experimental conditions. In this paper, we explore the use of the open-world tracking model for fish tracking and behavior analysis based on camera-based observation. We propose a unified framework that integrates fish position extraction and statistical behavior analysis to enable quantitative assessment of fish activities. The framework is validated with real-world fish recordings, where it successfully detects sudden behavioral changes caused by external disturbances. Experimental results demonstrate the potential of our approach to provide robust, scalable, and efficient fish behavior analysis. This work highlights the feasibility of leveraging an open-world tracking model for practical aquaculture monitoring applications.